Features in the primordial power spectrum? A frequentist analysis

TL;DR

This paper uses a frequentist statistical approach to assess whether observed features in the primordial power spectrum are statistically significant or likely due to noise, concluding that the data are consistent with a featureless spectrum.

Contribution

It introduces a Monte-Carlo based frequentist method to evaluate the significance of primordial spectrum features in CMB data.

Findings

26% of simulated datasets show greater likelihood improvement than actual data

Features in the spectrum are not statistically significant

WMAP data are consistent with a featureless power-law spectrum

Abstract

Features in the primordial power spectrum have been suggested as an explanation for glitches in the angular power spectrum of temperature anisotropies measured by the WMAP satellite. However, these glitches might just as well be artifacts of noise or cosmic variance. Using the effective Delta chi^2 between the best-fit power-law spectrum and a deconvolved primordial spectrum as a measure of "featureness" of the data, we perform a full Monte-Carlo analysis to address the question of how significant the recovered features are. We find that in 26% of the simulated data sets the reconstructed spectrum yields a greater improvement in the likelihood than for the actually observed data. While features cannot be categorically ruled out by this analysis, and the possibility remains that simple theoretical models which predict some of the observed features might stand up to rigorous statistical testing, our results suggest that WMAP data are consistent with the assumption of a featureless power-law primordial spectrum.